Académique Documents
Professionnel Documents
Culture Documents
n
D
i~1
D
i
S
t
i{
1
tt
i
f g
where 0,t
1
,t
2
,t
n
D
are the n
D
times when dividends are being paid, while for all
i51n
D
, the D
i
:
+
R
+
are continuous and monotone functions which represent the
amount of dividend (in cash) which is being paid at time t
i
, and they are thus
assumed to satisfy the constraint that D
i
(s)(s, for all s>0 and all i g{1, 2, , n
D
}.
The stochastic process {W
t
, t>0} is a standard Brownian motion on a filtered
probability space (V, F, F
t
t0
, ) with a risk-neutral probability measure Q and a
filtration F
t
t0
which satisfies the usual assumptions. The processes S and B are
thus completely specified under once their initial conditions S
0
and B
0
have been
specified.
We denote for all bounded Borel-sets E, by T
E
the class of stopping times (with
respect to the filtration F
t
t0
) which take values in E, i.e.
t[T
E
[ t[E ~1, $ : t $ t f g[F
t
Vt[E
For Lipschitz, polynomially bounded functions W:
+
R we can now define for all
s>0 the European and American Option price functions associated with the
contingent claim W and maturity T.0:
V
Eur
s ~B
0
W S
T
B
T
S
0
~s j
_ _
V
Am
s ~B
0
sup
t[T
0, T
W S
t
B
t
S
0
~s j
_ _
When n
D
50 (no dividends) good and fast approximations for these functions can
be found, for example using the binomial trees pioneered by Cox, Ross and
Rubinstein (1979). We define a more general discrete approximation process. Once
the initial value S
0
and the time of maturity T.0 have been fixed, we define for all
268 M. H. Vellekoop and J. W. Nieuwenhuis
n g
+
a time grid
D
n
~T=n,
n
~ kD
n
, k~0 . . . n f g
and a sequence of approximating discrete time processes S
n
, B
n
~ S
n
iD
n
, B
n
iD
n
_ _
i~0::n
where we take for every n an appropriate filtered probability space with risk-neutral
measure
n
and
S
n
iz1 D
n
~S
n
iD
n
J
n
iz1
, S
n
0
~S
0
B
n
iD
n
~e
{r
:
iD
n
with J
n
i
an F
iD
n
-measurable stochastic variable. The approximation to the European
and American Option price functions is
v
Eur
n
s ~B
0
n
W S
n
T
_ __
B
n
T
S
0
~s j
_
v
Am
n
s ~B
0
sup
t[T
0, Dn, 2Dn, ::, T f g
n
W S
n
t
_ __
B
n
t
S
0
~s j
_
Many possible choices for the distribution of J
n
i
under the corresponding discrete-
time martingale measure
n
exist. One can take a binomial or trinomial model or
any other stochastic processes which converges (weakly) to the BlackScholes model.
The most famous of these methods is the CRR-method of binomial trees, for which
we have the following result.
Lemma 1. Assume n
D
50 (i.e. there are no dividends) and that the payoff function is
that of a call or put, i.e. W(s)5(s2K)
+
or W(s)5(K2s)
+
for some K.0. For the model
defined above with
n
J
n
iz1
~u
n
_ _
~1{
n
J
n
iz1
~d
n
_ _
~
e
rD
n
{d
n
u
n
{d
n
, u
n
~
1
d
n
~e
s
D
n
p
we then have, for all s>0,
lim
n??
v
Eur
n
s ~V
Eur
s
lim
n??
v
Am
n
s ~V
Am
s
and this convergence is uniform on closed bounded sets.
Proof. See for example Amin and Khanna (1994), Lamberton (1998) and Leisen
(1998) for detailed proofs. &
In principle, extension of these methods to the case where cash dividends are being
paid is trivial. However, a direct approach often leads to non-recombining trees. For
example, in the CRR-method of the previous lemma, we have that if a dividend D
has been paid at timestep m, we have
S u
n
i
d
n
m{i
{D
_ _
d
n
= S u
n
i{1
d
n
m{iz1
{D
_ _
u
n
so one would have to build a new tree from every node after each dividend date
Efficient Pricing of Derivatives on Assets 269
which makes efficient computation for realistic problems impossible. We therefore
propose to make the tree recombining again by using an interpolation technique
after each dividend date. As an illustration, assume we have only one dividend date
t
D
(i.e. n
D
51 here), and define the set of nodes at the dividend date
A
n
~ s :
n
S
n
m n D
n
~s
_ _
> 0
_ _
with m(n) gN defined by the condition that
m n D
n
t
D
v m n z1 D
n
As before, we also have the set of nodes at maturity
A
n
~ s :
n
S
n
T
~s
_ _
> 0
_ _
:
We then suggest the following procedure.
Implementation
N Build a binomial tree with n time steps as usual. At the end of the tree we can
now calculate the value of W(s) for all s g
A
n
. We now work backwards
through the binomial tree as usual until we arrive at a dividend date (i.e. to
timestep m(n)). The result of having worked backwards through the tree is the
set of (approximate) values f
n
(s) for the option contract in all the points s of the
binomial tree at timestep m(n), i.e. for all s gA
n
. These values f
n
(s)
approximate the option values at a time just after the dividend has been paid,
given that the stock price at that time equals s.
N To be able to continue backwards through the tree we now have to find the
option values just before the dividend is paid. Since the stock price jumps down
with an amount D(S
t2
) when it goes ex-dividend, the option value for a stock
price s just before the dividend date equals the option value for a stock price
s2D(s) just after the dividend date. We thus need the option values f
n
in all the
points s2D(s) where s gA
n
, but we have only calculated the values of f
n
(s)
where s gA
n
. We therefore define a function which approximates the function
f
n
on the whole of
+
, based on the values of f
n
on A
n
. This interpolating
function is denoted by B
n
f
n , and we can then calculate the values B
n
f
n s{D s
as approximations for the values of f
n
(s2D(s)) with s gA
n
, which we need to
continue the binomial method in the point s.
N After that we work further backwards through the tree, until we reach time
zero, and the option value has been found.
The extension of this method to more than one dividend is obvious. Also note that
it is not essential to use a binomial tree, but in the numerical examples given later, we
will always use binomial trees.
It is clear that the convergence of this approximating procedure to the correct
value when the number of time steps n goes to infinity, will depend on the
approximation procedure B that we use in the algorithm. Even when the quality of
the approximation becomes better as n increases, we cannot immediately conclude
270 M. H. Vellekoop and J. W. Nieuwenhuis
that our option price approximations will converge to the correct values when n goes
to infinity, because (1) our interpolation is based on values which still contain an
error (i.e. the values {f
n
(s), s gA
n
} that we use to interpolate are only an
approximation of the true option values in those points); and (2) the points A
n
in
between which we approximate change as a function of n, and fill out the whole of
+
only in the limit.
Because of this, we will need to pose explicit conditions on our interpolation
procedure in the next section, to be able to prove there that convergence does indeed
take place under the restrictions posed. This will be the subject of the next section.
3. Formal Proof of Convergence
We define the following option price functions that we would like to approximate
V
Eur
s ~B
0
W S
T
B
T
S
0
~s j
_ _
V
Am
s ~B
0
sup
t[T
0, T
W S
t
B
t
S
0
~s j
_ _
with the dynamics of S
t
under defined as in (1), but with only one dividend, i.e.
n
D
51 (the more general case can be handled similarly). We first define two functions
which represent the pricing functions directly after the dividend has been paid
f
Eur
s ~B
t
D
W S
T
B
T
S
t
D
~s j
_ _
f
Am
s ~B
t
D
sup
t[T
t
D
, T
W S
t
B
t
S
t
D
~s j
_ _
Lemma 2. Under the assumptions stated above we have
V
Eur
s ~B
0
f
Eur
S
t
D
{
{D S
t
D
{
B
t
D
S
0
~s j
_ _
V
Am
s ~B
0
sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
f
Am
S
t
D
{
{D S
t
D
{
B
t
D
S
0
~s j
_ _
Proof. The European case is trivial, since
W S
T
B
T
S
0
~s j
_ _
~
W S
T
B
T
S
t
D
j
_ _
S
0
~s j
_ _
~
f
Eur
S
t
D
B
t
D
S
0
~s j
_ _
~
f
Eur
S
t
D
{
{D S
t
D
{
B
t
D
S
0
~s j
_ _
Efficient Pricing of Derivatives on Assets 271
For the American case, we note that
V
Am
s
_
B
0
~ sup
t[T
0, T
W S
t
B
t
1
tvt
D
f g
z
W S
t
B
t
1
tt
D
f g
S
0
~s
_ _
~ sup
t[T
0, T
W S
t
B
t
1
tvt
D
f g
z1
tt
D
f g
W S
t
B
t
F
t
D
j
_ _
S
0
~s
_ _
sup
t[T
0, T
W S
t
B
t
1
tvt
D
f g
z1
tt
D
f g
sup
f[T
t
D
, T
W S
f
B
f
S
t
D
j
_ _
S
0
~s
_
_
_
_
but the last inequality is in fact an equality. To see this, assume that there exist a
t
[T
0, T
and f
[T
t
D
, T
such that
W S
t
B
t
1
t
vt
D
f g
z1
t
t
D
f g
W S
f
_ _
B
f
S
t
D
j
_ _
S
0
~s
_ _
> sup
t[T
0, T
W S
t
B
t
1
tvt
D
f g
z1
tt
D
f g
W S
t
B
t
F
t
D
j
_ _
S
0
~s
_ _ 3
then we can define the random variable
tt~t
1
t
vt
D
f g
zf
1
t
t
D
f g
for which one can easily check that it is a stopping time in T
0, T
as well, so the right-
hand side of (3) must be equal to or larger than
W S~
t
B~
t
1 ~
tvt
D
f g
z1 ~
tt
D
f g
W S~
t
B~
t
F
t
D
j
_ _
S
0
~s
_ _
~
W S
t
B
t
1
t
vt
D
f g
z1
t
t
D
f g
W S
f
_ _
B
f
S
t
D
j
_ _
S
0
~s
_ _
which is the left-hand side of (3), which would be a contradiction. We thus conclude
that
V
Am
s
_
B
0
~ sup
t[T
0, T
W S
t
B
t
1
tvt
D
f g
z1
tt
D
f g
sup
f[T
t
D
, T
W S
f
B
f
S
t
D
j
_ _
S
0
~s
_
_
_
_
and using the fact that if t is a stopping time, so is tt
D
, we find
272 M. H. Vellekoop and J. W. Nieuwenhuis
V
Am
s
_
B
0
~ sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
sup
f[T
t
D
, T
W S
f
B
f
S
t
D
j
_ _
S
0
~s
_
_
_
_
~ sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z
f
Am
S
t
D
B
t
D
1
t~t
D
f g
S
0
~s
_ _
~ sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z
f
Am
S
t
D
{
{D S
t
D
{
B
t
D
1
t~t
D
f g
S
0
~s
_ _
and this proves the result. &
For all functions g: A
n
R we need to define an approximating function which
defines values for points in the set
A
n
~ a{D a , a[A
n
f g, and we will denote these
values by B
n
g
s for s[
A
n
. We will do this in a relatively simple way, by using linear
interpolation on the values of
A
n
augmented by the value in S50, which will, for
example, be zero for a call and equal to the strike K for theput, and will be a known
constant for all other types of options as well. This guarantees that we can indeed use
interpolation for all values in
A
n
and do not need to extrapolate.
In contrast to the case where we had no dividends, we now use the following
approximations, as suggested by the previous lemma
V
Eur
n
s ~B
n
0
n
B
n
f
Eur
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
n
t
D
S
0
~s j
_
_
_
_
V
Am
n
s ~B
n
0
sup
t[T
0, Dn, ::, t
n
D
f g
n W S
n
t
_ _
B
t
1
tvt
D
f g
z1
t~t
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
D
S
0
~s j
_
_
_
_
with the functions f
Eur
n
, f
Am
n
defined by
f
Eur
n
s ~B
n
t
n
D
n
W S
n
T
_ __
B
n
T
S
t
n
D
~s
_ _
f
Am
n
s ~B
n
t
n
D
sup
t[T
t
n
D
, ..., T
f g
n
W S
n
t
_ __
B
n
t
S
t
n
D
~s
_ _
and
t
n
D
~m n D
n
, t
n
D{
~ m n {1 D
n
We will use the following assumptions:
N (A1) For all n, the stochastic process S
n
iD
n
_
B
n
iD
n
is Markov, it is a
n
-
martingale, i.e. for all i g{0, 1, , n}
Efficient Pricing of Derivatives on Assets 273
n
S
n
iz1 D
n
B
n
iz1 D
n
S
n
iD
n
B
n
iD
n
_ _
~
S
n
iD
n
B
n
iD
n
4
and the sequence of stochastic variables S
n
t
n
D{
, n[
z
_ _
is uniformly integrable
under
n
, i.e.
lim
M??
sup
n
n
S
n
t
n
D{
1
S
n
t
n
D{
M
_ _
_
_
_
_
~0
N (A2) The function W:
+
R satisfies a uniform Lipschitz condition, i.e.
A h
W
> 0 Vx, y [
z
W x {W y j jh
W
x{y j j
and its absolute value is bounded by a polynomial.
N (A3) The approximating sequence of discrete time option prices is such that in
the absence of dividends (i.e. when the function D equals zero on its entire
domain) we have uniform convergence (on closed, bounded sets) to the correct
values, i.e. the conclusion of Lemma 1 holds for all payoff functions W
satisfying condition (A2).
It is well known that condition (A3) is satisfied for standard calls and puts in the
CRR-model. The hard part to prove is uniform convergence for the American Put,
see for example the paper by Lamberton (1998) for a proof. The European case
(which includes the American Call if there are no dividends) is relatively easy.
Condition (A1) is certainly satisfied in the CRR-model, since the process S
n
converges weakly to the continuous process S in that case.
We will first prove one more lemma that will be needed in the proof for our main
theorem.
Lemma 3. Under conditions (A1) and (A2) we have that f
Am
and f
Eur
are uniformly
Lipschitz, and so are the functions f
Am
n
and f
Eur
n
for all possible n g
+
.
Proof. As an example, we prove the result for the functions f
Am
n
, all other cases can
be handled similarly. Define the stochastic variable G(s
1
, s
2
, t) for s
1
, s
2
g
+
and
k g{m(n), , n} as
G s
1
, s
2
, k ~B
n
t
n
D
W s
1
S
n
kD
n
S
n
t
n
D
_ _
{W s
2
S
n
kD
n
S
n
t
n
D
_ _
B
n
kD
n
then for all stopping time f[T
0, D
n
, ::, T f g
, by the optional stopping theorem,
equation(4)
274 M. H. Vellekoop and J. W. Nieuwenhuis
n
G s
1
, s
2
, f j j h
W
s
1
{s
2
j jB
n
t
n
D
n
S
n
f
S
n
t
n
D
B
n
fD
n
~h
W
s
1
{s
2
j j
so since for all s and s
1
f
Am
n
s ~B
n
t
n
D
sup
t[T
t
n
D
, ..., T
f g
n
W s
S
n
t
S
n
t
n
D
_ __
B
n
t
_ _
~ sup
t[T
t
n
D
, ..., T
f g
n
G s, s
1
, t zB
n
t
n
D
W s
1
S
n
t
S
n
t
n
D
_ __
B
n
t
_ _
we have
f
Am
n
s B
n
t
n
D
sup
t[T
t
n
D
, ..., T
f g
n
W s
1
S
n
t
S
n
t
n
D
_ __
B
n
t
_ _
z sup
t[T
t
n
D
, ..., T
f g
n
G s, s
1
, t j j
f
Am
n
s B
n
t
n
D
sup
t[T
t
n
D
, ..., T
f g
n
W s
1
S
n
t
S
n
t
n
D
_ __
B
n
t
_ _
{ sup
t[T
t
n
D
, ..., T
f g
n
G s, s
1
, t j j
so
f
Am
n
s {f
Am
n
s
1
h
W
s{s
1
j j
which proves the result. &
We can now prove our main result of the section.
Theorem 1. Under the assumptions (A1),(A2) and (A3), we have that for all s g
+
lim
n??
V
Eur
n
s ~V
Eur
s
lim
n??
V
Am
n
s ~V
Am
s
Proof. We will prove the American case; the European case is simpler and can be
proven analogously. We have since B
n
0
~B
0
Efficient Pricing of Derivatives on Assets 275
1
B
0
V
Am
n
s {V
Am
s
_ _
~ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
n
t
n
D
S
0
~s j
_
_
_
_
{ sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
f
Am
S
t
D
{
{D S
t
D
{
B
t
D
S
0
~s j
_ _
~E
n
approx
s zE
n
disc
s
where
E
n
approx
s ~ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
t~t
n
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
{ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
t~t
n
D
f g
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
E
n
disc
s ~ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
t~t
n
D
f g
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
{ sup
t[T
0, t
D
W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
f
Am
S
t
D
{
{D S
t
D
{
B
t
D
S
0
~s j
_ _
Now for all s.0, E
n
disc
s converges to zero due to assumption (A3), since the
requirement that all European and American option values converge for the case
without dividends, allows us to apply the results of Amin and Khanna (1994), which
show that we even have convergence for payoff functions which are time-varying (as
they are in the expression above), since the uniform integrability condition on a time-
varying payoff process W(S
t
, t) which is needed for their result to hold true is satisfied
because of our condition (A2). Hence, we just need to consider E
n
approx
s .
Since both f
Am
and f
Am
n
are uniformly Lipschitz by Lemma 3 and since this
property is inherited by linear interpolation, we can choose a
SS > 0 and G.0 with
s >
SS[ B
n
f
Am
n
s{D s
z f
Am
s{D s
vG s{D s j jG s j j 5
for all n, where we used the fact that s2D(s) was assumed to be positive. We can
then, due to our condition (A1), choose for every e.0 an S
e
>
SS so large that
276 M. H. Vellekoop and J. W. Nieuwenhuis
Vn[
z
n
S
t
n
D{
1
S
n
t
D{
>S
e
_ _
S
0
~s j
_ _
v
e
2G
e
{rT
6
Write
E
n
approx
s ~ sup
t[T
0, Dn, ::, t
n
D
f g
n
E
n
1
t S
0
~s j
_
{ sup
t[T
0, Dn, ::, t
n
D
f g
n
E
n
2
t S
0
~s j
_
where
E
n
1
t ~
W S
t
B
t
1
tvt
n
D
f g
z1
t~t
n
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
E
n
2
t ~
W S
t
B
t
1
tvt
D
f g
z1
t~t
D
f g
f
Am
S
t
D
{
{D S
t
D
{
B
t
D
and by (5),
E
n
1
t
W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
zG S
t
n
D{
1
S
t
n
D{
>
~
S
e
_ _
E
n
2
t
W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
{G S
t
n
D{
1
S
t
n
D{
>
~
S
e
_ _
and together with (6) this shows that
E
n
approx
s
ez
sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
B
n
f
Am
n
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
{ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
~0 7
because if
Vd > 0 AR
d
[
z
Vn > R
d
sup
0s
~
S
e
B
n
f
Am
n
s {f
Am
s
vd
Efficient Pricing of Derivatives on Assets 277
then we we have that for n.R
d
E
n
approx
ez
sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
dz
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
_
_
_
_
S
0
~s j
_
_
_
_
{ sup
t[T
0, Dn, ::, t
n
D
f g
n W S
t
B
t
1
tvt
n
D
f g
z1
S
t
n
D{
~
S
e
_ _
1
t~t
n
D
f g
f
Am
S
n
t
n
D
{
{D S
n
t
n
D
{
_ _ _ _
B
t
n
D
S
0
~s j
_
_
_
_
dze
for d and e arbitrarily small. To prove (7), take n.N
e
so large that max A
n
>
SS
e
(this is possible since max(A
n
)R for nR because otherwise (A3) will clearly not
hold for certain choices of W). We set
a
n
s ~arg min
x[A
n
x{s j j
and write
B
n
f
Am
n
s {f
Am
s
B
n
f
Am
n
s {B
n
f
Am
n
a
n
s
z B
n
f
Am
n
a
n
s {f
Am
n
a
n
s
z f
Am
n
a
n
s {f
Am
a
n
s
z f
Am
a
n
s {f
Am
s
the second term is zero by definition and we can bound the first and last term using
the uniform Lipschitz properties that we derived before equation(6) for both B
n
f
Am
n
s
and f
Am
(say the values of the Lipschitz constants of the functions do not exceed
H.0), so
B
n
f
Am
n
s {f
Am
s
2H
:
mesh A
n
z f
Am
n
a
n
s {f
Am
a
n
s
8
where mesh A
n
~max A
n
1
{0, max
k~1::n{1
A
n
kz1
{A
n
k
_ _
_ _
converges to zero for nR,
since otherwise (A3) would clearly not hold. We now write
sup
0s
~
S
e
B
n
f
Am
n
s {f
Am
s
2H
:
mesh A
n
z sup
0s
~
S
e
f
Am
n
a
n
s {f
Am
a
n
s
~0:
This proves (7) and hence the result. &
278 M. H. Vellekoop and J. W. Nieuwenhuis
Remarks
N In practical cases, where one can obviously use only a finite number of time
steps, it is important to make sure that the approximation step is accurate
enough, i.e. that there are enough nodes in the tree at the first dividend date.
N Note that for convex pricing functions, the linear interpolation step always
overestimates the value that it tries to approximate.
N One could use a cubic spline approximation instead of a linear approximation.
To prove convergence we needed a Lipschitz condition for the tail in
Equation 6 and in the error bound in Equation 8. A cubic spline approximation
will not automatically inherit the Lipschitz property of the function it
approximates, so the proof given does not go through directly. However, if one
applies a cubic spline method which inherits monotonicity from the function it
approximates, then the proof can easily be shown to go through if the price
functions f
n
and f are monotone (which is clearly the case for standard Call and
Put contracts). Such spline methods exist (Kuijt and VanDamme, 1999).
N We have only given the proof for the case of one dividend, but the extension to
more dividends is obvious. Note however, that one has to check that the
conditions (A1), (A2), and (A3) remain valid for every single step, which may
be complicated for exotic payoff functions, since little is known in general
about the rates of convergence for such cases.
4. Numerical Results
To illustrate results, the proposed method and other available methods were tested
on a range of European and American options. A standard binomial tree and linear
interpolation were used in all cases described below.
In a first experiment, we price an American call with time to maturity 1 year and a
single dividend of value 7.0 at time t
1
, where t
1
was taken to be 0.1, 0.5 and 0.9 in
different experiments. As is well known, the exact value of this American option can
be calculated using a single integral. The volatility and interest rate were taken to be
s530% and r55% respectively, and the current stock price was S5100. We took
strike prices equal to 70, 100 and 130. Note that quite a long time to maturity and
high dividends were taken to make differences between the pricing methods more
transparent.
Table 1 shows the exact value, the results of our algorithm (VN) for 250, 500 and
1000 steps in the binomial tree respectively, and the results for the two other
available methods to deal with American options with a single dividend, the Roll
GeskeWhaley method (RGW) and the Black approximation (see Roll, 1977;
Whaley, 1981; Geske, 1979; Black, 1975) for more information on these methods).
We also give the results (VNRE) of Richardson extrapolation on our method (where
the maximal number of steps used was 64000) to show that we have convergence.
Note that the results are very satisfactory for a more moderate number of steps,
while the two other models may significantly misprice the options.
In the second experiment, we considered a European call with a time to maturity
of 7 years and cash dividends equal to 6.0, 6.5, 7.0, 7.5, 8.0, 8.0, 8.0, in consecutive
Efficient Pricing of Derivatives on Assets 279
years. The time between dividend dates was taken to be one year, and the time of the
first dividend was again varied: we present results for t
1
50.1, t
1
50.5 and t
1
50.9 with
s525%, r56% and S5100.
The results are shown in Table 2 for our methods with 250, 500 and 1000 time
steps, and alternative methods formulated by Haug, Haug and Lewis (2003, HHL),
Bos, Gairat and Shepelva (2003, BGS), Bos and Vandermark (2002, BvdM) and
Beneder and Vorst (2002, BV). The exact details of their methods can be found in the
corresponding papers in the references.
As we can see, the last three methods show substantial differences when compared
to our method. The approximation method of Haug, Haug and Lewis (who replace a
multiple integration by a succession of single integrations over BlackScholes-like
approximating functions) performs extremely well. In fact it outperforms the other
existing methods in all cases and outperforms the proposed method when only 250
steps are used. When our method uses more steps, it gives results closer to the correct
value (as expected, since we actually proved convergence). However, it should be
noted that the last three methods in the table are obviously quicker to calculate since
they are all based on the BlackScholes formula for European options in which an
adjusted volatility is inserted, so there is a trade-off in accuracy and time involved
here.
In the last experiment we took the same options as before, but this time looked at
the American call option. None of the other methods mentioned above can deal with
American options with multiple dividends and no exact formula exists. We again
note that finite difference methods can also be used to price American options in the
presence of multiple cash dividends, but to be sure that there is convergence to the
correct value, a detailed analysis similar to ours would be needed.
Table 1. American call, s530%, r55%, S5100
t
1
K Exact VNRE VN1000 VN500 VN250 RGW Black
0.1 70 30.38 30.38 30.38 30.37 30.35 30.37 30.35
100 10.29 10.29 10.29 10.29 10.28 10.20 10.20
130 3.00 3.00 3.00 3.00 2.99 2.93 2.93
0.5 70 32.13 32.13 32.12 32.11 32.10 32.00 31.98
100 11.33 11.33 11.32 11.31 11.30 10.86 10.28
130 3.28 3.28 3.28 3.28 3.28 2.98 2.96
0.9 70 33.92 33.92 33.91 33.90 33.88 33.66 33.91
100 13.49 13.49 13.48 13.47 13.43 12.75 13.40
130 4.17 4.17 4.16 4.15 4.14 3.62 4.04
0.1 70 0.00% 20.03% 20.06% 20.11% 20.04% 20.12%
100 0.00% 0.00% 0.00% 20.05% 20.84% 20.84%
130 0.00% 0.04% 0.14% 20.04% 22.07% 22.07%
0.5 70 0.00% 20.03% 20.06% 20.10% 20.40% 20.48%
100 0.00% 20.06% 20.12% 20.27% 24.13% 29.26%
130 0.00% 0.01% 0.05% 20.01% 29.03% 29.66%
0.9 70 0.00% 20.03% 20.06% 20.10% 20.76% 20.03%
100 0.00% 20.11% 20.17% 20.46% 25.49% 20.64%
130 0.00% 20.17% 20.36% 20.63% 213.28% 23.09%
280 M. H. Vellekoop and J. W. Nieuwenhuis
We therefore again tried to find its value by applying the proposed method, using
a very large number of steps (64000) and Richardson extrapolation, and compared
this value to the same method using a more modest numbers of steps and to the
results of the Longstaff and Schwartz (2001) method of Monte Carlo simulation for
American options. The results can be found in Table 3. We see that for 1000 steps we
already have very satisfactory results but for a smaller number of steps we may run
into problems when the first dividend comes early, since the approximation of the
option values at that point is then based on too few interpolation points. As
mentioned in the preceding section, more time steps are needed to obtain good
results in this case.
Notice that the results of the Monte Carlo method of Longstaff and Schwarz are
close to ours. The Monte Carlo estimates are based on 2 000 000 antithetic samples
and three basis functions in the LongstaffSchwartz least squares approximation.
The maximal absolute error of 1% we observe for the Monte Carlo prices is not
exceptional: we also ran the Monte Carlo simulation for the setup of Table 1 (where
we know the exact value of the option) and found a maximal absolute error of 2%
there.
Finally, in Table 4 we took the same setup as in the previous experiment, but we
now took all cash dividends equal to 3, and we removed the first and last dividend
dates. This ensures that for all cases considered here we have that
DvK 1{e
{r t
iz1
D
{t
i
D
_ _
for i506, where {t
1
D
,t
2
D
, . . . ,t
6
D
} are the dividend dates and t
0
D
~0 and t
7
D
~T. This
condition guarantees that it is never optimal to exercise the American option and we
Table 2. European call, s525%, r56%, S5100
t
1
K VNRE VN1000 VN500 VN250 HHL BGS BvdM BV
0.1 70 24.90 24.92 24.98 25.24 25.05 24.71 24.74 23.43
100 17.43 17.46 17.51 17.74 17.50 17.42 17.08 16.41
130 12.40 12.43 12.47 12.69 12.40 12.50 11.94 11.83
0.5 70 26.08 26.10 26.10 26.14 26.20 25.87 25.94 24.58
100 18.48 18.50 18.51 18.56 18.51 18.45 18.15 17.51
130 13.29 13.31 13.33 13.40 13.24 13.38 12.84 12.83
0.9 70 27.21 27.23 27.23 27.28 27.30 26.99 27.10 25.67
100 19.48 19.50 19.52 19.55 19.48 19.43 19.19 18.54
130 14.13 14.16 14.17 14.25 14.06 14.21 13.73 13.77
0.1 70 0.10% 0.32% 1.39% 0.62% 20.74% 20.64% 25.89%
100 0.16% 0.43% 1.73% 0.39% 20.11% 22.01% 25.86%
130 0.27% 0.59% 2.30% 20.03% 0.81% 23.69% 24.56%
0.5 70 0.07% 0.08% 0.24% 0.44% 20.80% 20.54% 25.75%
100 0.12% 0.17% 0.41% 0.15% 20.19% 21.78% 25.28%
130 0.22% 0.30% 0.88% 20.32% 0.71% 23.32% 23.45%
0.9 70 0.07% 0.08% 0.23% 0.31% 20.83% 20.41% 25.68%
100 0.10% 0.17% 0.37% 20.01% 20.27% 21.50% 24.85%
130 0.21% 0.28% 0.87% 20.47% 0.57% 22.86% 22.58%
Efficient Pricing of Derivatives on Assets 281
should therefore find exactly the same prices when we use the European or the
American version of our algorithm. We see in Table 4 that the prices are indeed
exactly the same. We also find excellent agreement with Monte Carlo estimates,
Table 4: American Call, s525%, r56%, S5100
t
1
K VN5000 Am VN5000 Eur Monte Carlo
0.1 70 45.873 45.873 45.878
100 33.724 33.724 33.722
130 24.868 24.868 24.857
0.5 70 46.238 46.238 46.232
100 34.089 34.089 34.085
130 25.206 25.206 25.195
70 46.594 46.594 46.591
0.9 100 34.443 34.443 34.440
130 25.532 25.532 25.533
0.1 70 0.000% 0.010%
100 0.000% 20.006%
130 0.000% 20.045%
0.5 70 0.000% 20.013%
100 0.000% 20.012%
130 0.000% 20.045%
70 0.000% 20.007%
0.9 100 0.000% 20.009%
130 0.000% 0.004%
Table 3. American call, s525%, r56%, S5100
t
1
K VNRE VN1000 VN500 VN250 LonSch
0.1 70 31.14 31.06 31.07 31.25 30.91
100 18.32 18.32 18.35 18.54 18.17
130 12.48 12.51 12.54 12.73 12.57
0.5 70 33.47 33.40 33.37 33.27 33.30
100 20.04 20.04 20.02 20.00 20.07
130 13.75 13.76 13.75 13.78 13.80
0.9 70 35.52 35.48 35.42 35.31 35.28
100 21.86 21.85 21.82 21.78 21.65
130 15.21 15.21 15.19 15.21 15.09
0.1 70 20.24% 20.22% 0.36% 20.69%
100 0.03% 0.16% 1.18% 20.80%
130 0.20% 0.45% 2.02% 0.71%
0.5 70 20.19% 20.29% 20.58% 20.45%
100 20.03% 20.14% 20.25% 0.15%
130 0.10% 0.00% 0.26% 0.36%
0.9 70 20.13% 20.30% 20.60% 20.66%
100 20.07% 20.18% 20.35% 20.96%
130 0.01% 20.11% 20.02% 20.82%
282 M. H. Vellekoop and J. W. Nieuwenhuis
which were generated using 2000000 antithetic samples and with control variates
consisting of European options with the same parameters but without dividends.
5. Conclusions
We have presented an efficient method to deal with cash dividends in equity option
pricing methods, under the assumption that in between dividend dates the asset
follows lognormal dynamics, and where the same dynamics are used to price all
derivative products. The interpolation method used was shown to converge under an
additional assumption of uniform convergence for approximation methods without
dividends, since this allows one to use a conditioning argument over the dividend
dates.
Our method does not give closed-form formulas for European options (such as the
BenederVorst, the BosVdMark and the BosGairatShepeleva models) but it has
the advantage that it can also be used for American options with multiple dividends.
Since the interpolation step takes only a modest amount of time, the speed of the
proposed method is roughly the same as that of the underlying binomial, trinomial
or other schemes without dividends on which our method can be based. We therefore
expect it to be useful for consistent pricing of derivatives on assets which involve cash
dividends.
Acknowledgements
This research was partially funded by The Derivatives Technology Foundation.
Their support is gratefully acknowledged.
References
Amin, K. and Khanna, A. (1994) Convergence of American option values from discrete- to continuous-
time financial models, Mathematical Finance, 4, pp. 289304.
Beneder, R. and Vorst, T. (2002) Options on dividend paying stocks, in: J. Yong (Ed.) Recent Develop-
ments in Mathematical finance (Shanghai, 2001) (River Edge, NJ: World Scientific Publishing).
Black, F. (1975) Fact and fantasy in the use of options, Financial Analysts Journal, pp. 3672.
Black, F. and Scholes, M. (1973) The pricing of options and corporate liabilities, Journal of Political
Economy, 81, pp. 637654.
Bos, R. et al. (2003) Dealing with discrete dividends, Risk Magazine, 16, pp. 109112.
Bos, R. and Vandermark, S. (2002) Finessing fixed dividends, Risk Magazine, 15, pp. 157158.
Cox, J. C. et al. (1979) Option pricing: a simplified approach, Journal of Financial Economics, 7, pp.
229263.
Frishling, V. (2002) A discrete question, Risk Magazine, 15, pp. 115116.
Geske, R. (1979) A note on an analytical formula for unprotected american call options on stocks with
known dividends, Journal of Financial Economics, 7, pp. 375380.
Haug, E. G. et al. (2003) Back to basics: a new approach to the discrete dividend problem, Wilmott
Magazine, pp. 3747.
Kuijt, F. and VanDamme, R. (1999) Monotonicity preserving interpolatory subdivision schemes, Journal
of Computational & Applied Mathematics, 101, pp. 203229.
Lamberton, D. (1998) Error estimates for the binomial approximation of american put options, Annals of
Applied Probability, 8, pp. 206233.
Leisen, D. P. (1998) Pricing the american put option: a detailed convergence analysis for binomial models,
Journal Economics Dynamics and Control, 22, pp. 14191444.
Efficient Pricing of Derivatives on Assets 283
Longstaff, F. A. and Schwartz, E. S. (2001) Valuing American options by simulation: a simple least-
squares method, Review of Financial Studies, 14, pp. 113147.
Merton, R. C. (1973) Theory of rational option pricing, Bell Journal of Economics and Management
Science, 4, pp. 141183.
Roll, R. (1977) An analytical formula for unprotected american call options on stocks with known
dividends, Journal of Financial Economics, 5, pp. 251258.
Whaley, R. E. (1981) On the valuation of american call options on stocks with known dividends, Journal
of Financial Economics, 9, pp. 207211.
Wilmott, P. et al. (1993) Option Pricing: Mathematical Models and Computation (Oxford: Oxford
Financial Press).
284 M. H. Vellekoop and J. W. Nieuwenhuis